96 Using text mining and graph analytics to optimally identify related biomedical research publications. (23rd February 2023)

• Record Type: Journal Article
• Authors: Booth, John; Eriksson, Maria; Bowyer, Stuart A; Briggs, Lydia; Bryant, William A; Key, Daniel; Shah, Mohsin; Spiridou, Anastassia; Sebire, Neil J
• Is Part Of: Archives of disease in childhood. Volume 108(2023)Supplement 1
• Page Start: A36
• Page End: A36
• Publication Date: 2023-02-23
• Subjects: Children -- Diseases -- Periodicals; Infants -- Diseases -- Periodicals; 618.920005
• DOI: 10.1136/archdischild-2023-gosh.96
• Languages: English
• ISSNs: 0003-9888

Abstract

Abstract : Background: There are over 30, 000 scientific journals with close to two million articles being published each year. We explore how text mining and graph analytics can be used to streamline the process of identifying relevant papers within a specific subject area making the process more objective and reducing bias. Methods: A broad search criterion deployed in PubMed, IEEE and ACM search engines returns a set of titles and abstracts. Text mining routines are used to split the abstract into sentences and then Named Entity Recognition plus Linking to the Universal Medical Language System (UMLS) ontology (NER+L) applied to each sentence to identify clinical concepts. A graph was created for all concepts occurring in the same sentence and then the concepts were ranked using eigenvector centrality scores. The overall period covering all abstracts was then split into several sub-periods and for each sub-period graphs created, and the concepts ranked. Results: A search for epilepsy treatment in children returned 34k abstracts over a period from 1950 to-date. The abstracts were sub-divided into 12 sub-periods including 2020-21 and 2022. Having created graphs for all abstracts and each sub-period, a common set of concepts across all periods was identified these were then abstracted from the sub-period ranked lists. The ranked concepts for 2020-21 identified 'COVID-19' and 'lockdown' as being newly used concepts. The rankings for 2022 identified new genes and medications which are being researched, as well as indicating which medications are falling in research interest. Conclusions: This study demonstrates the feasibility of using text mining and graph analytics to objectively identify papers for manually review given a broad search criterion. This approach is both efficient and reduces biasand is applicable to any domain/clinical area without requiring an extensive domain knowledge.